Landform mapping is the initial step of many geomorphological analyses(e.g.assessment of natural hazards and natural resources)and requires vast resources to be applied to wide areas at high-resolution.Among geomorpho...Landform mapping is the initial step of many geomorphological analyses(e.g.assessment of natural hazards and natural resources)and requires vast resources to be applied to wide areas at high-resolution.Among geomorphological objects,we focus on glacial moraine mapping,since it is a task relevant to many fields(e.g.paleoclimate and glacial geomorphology).Here we proposed to exploit the potential of Deep Learning-based approaches to map moraine landforms by exploiting multi-source remote sensing imagery.To this end,we propose the first Deep Learning model to map glacial moraines,namely MorNet.As multi-source remote sensing information,we combine together three different sources:Topographic(Pleiades-derived DSM),Multispectral(Sentinel-2),and SAR(Sentinel-1)data.To cope with such heterogeneous information,the proposed model has a dedicated branch for each input source and,a late fusion mechanism is leveraged to combine them with the aim to provide the final mapping.The performance of the MorNet model is evaluated on several glacier valleys in China in the Himalayan range.This area contains minimally eroded moraines,so they are well-defined and of varied morphology.The behavior of the proposed method is compared to models using individual mono-source models in order to highlight the benefit to simultaneously leverage multi-source information.The use of multi-source data allows MorNet to exploit the complementarity of the three input sources and improve its performance from an f1-score of about 41.6 using a single source to 52.8 using three sources.MorNet provides a first-order moraine map through its ability to identify well-defined moraines.Consequently,MorNet can identify areas likely to contain moraines and intends to be used as a tool by experts to facilitate and support large-scale mapping.展开更多
Event-based surveillance systems are at the crossroads of human and animal(and plant and ecosystem)health,epidemiology,statistics,and informatics.Thus,their deployment faces many challenges specific to each domain and...Event-based surveillance systems are at the crossroads of human and animal(and plant and ecosystem)health,epidemiology,statistics,and informatics.Thus,their deployment faces many challenges specific to each domain and their intersections,such as relations among automation,artificial intelligence,and expertise.In this context,ourwork pertins to the extraction of epidemiological events in textual data(i.e.news)by unsupervised methods.We define the event extraction task as detecting pairs of epidemiological entities(e.g.a disease name and location).The quality of the ranked lists of pairs was evaluated using specific ranking evaluation metrics.We used a publicly available annotated corpus of 438 documents(i.e.news articles)related to animal disease events.The statistical approach was able to detect event-related pairs of epidemiological features with a good trade-off between precision and recall.Our results showed that using a window of words outperformed document-based and sentence-based approaches,while reducing the probability of detecting false pairs.Our results indicated that Mutual Information was less adapted than the Dice coefficient for ranking pairs of features in the event extraction framework.We believe that Mutual Information would be more relevant for rare pair detection(i.e.weak signals),but requires higher manual curation to avoid false positive extraction pairs.Moreover,generalising the country-level spatial features enabled better discrimination(i.e.ranking)of relevant disease-location pairs for event extraction.展开更多
Several internet-based surveillance systems have been created to monitor the web for animal health surveillance.These systems collect a large amount of news dealing with outbreaks related to animal diseases.Automatica...Several internet-based surveillance systems have been created to monitor the web for animal health surveillance.These systems collect a large amount of news dealing with outbreaks related to animal diseases.Automatically identifying news articles that describe the same outbreak event is a key step to quickly detect relevant epidemiological information while alleviating manual curation of news content.This paper addresses the task of retrieving news articles that are related in epidemiological terms.We tackle this issue using text mining and feature fusion methods.The main objective of this paper is to identify a textual representation in which two articles that share the same epidemiological content are close.We compared two types of representations(i.e.,features)to represent the documents:(i)morphosyntactic features(i.e.,selection and transformation of all terms from the news,based on classical textual processing steps)and(ii)lexicosemantic features(i.e.,selection,transformation and fusion of epidemiological terms including diseases,hosts,locations and dates).We compared two types of term weighing(i.e.,Boolean and TF-IDF)for both representations.To combine and transform lexicosemantic features,we compared two data fusion techniques(i.e.,early fusion and late fusion)and the effect of features generalisation,while evaluating the relative importance of each type of feature.We conducted our analysis using a corpus composed of a subset of news articles in English related to animal disease outbreaks.Our results showed that the combination of relevant lexicosemantic(epidemiological)features using fusion methods improves classical morphosyntactic representation in the context of disease-related news retrieval.The lexicosemantic representation based on TF-IDF and feature generalisation(F-measure=0.92,r-precision=0.58)outperformed the morphosyntactic representation(F-measure=0.89,r-precision=0.45),while reducing the features space.Converting the features into lower granular features(i.e.,generalisation)contributed to improving the results of the lexicosemantic representation.Our results showed no difference between the early and late fusion approaches.Temporal features performed poorly on their own.Conversely,spatial features were the most discriminative features,highlighting the need for robust methods for spatial entity extraction,disambiguation and representation in internet-based surveillance systems.展开更多
基金funded by the University of Montpellier(IR's PhD fellowship)and the French Centre National d'Etudes Spatiales(access to Pleiades images).
文摘Landform mapping is the initial step of many geomorphological analyses(e.g.assessment of natural hazards and natural resources)and requires vast resources to be applied to wide areas at high-resolution.Among geomorphological objects,we focus on glacial moraine mapping,since it is a task relevant to many fields(e.g.paleoclimate and glacial geomorphology).Here we proposed to exploit the potential of Deep Learning-based approaches to map moraine landforms by exploiting multi-source remote sensing imagery.To this end,we propose the first Deep Learning model to map glacial moraines,namely MorNet.As multi-source remote sensing information,we combine together three different sources:Topographic(Pleiades-derived DSM),Multispectral(Sentinel-2),and SAR(Sentinel-1)data.To cope with such heterogeneous information,the proposed model has a dedicated branch for each input source and,a late fusion mechanism is leveraged to combine them with the aim to provide the final mapping.The performance of the MorNet model is evaluated on several glacier valleys in China in the Himalayan range.This area contains minimally eroded moraines,so they are well-defined and of varied morphology.The behavior of the proposed method is compared to models using individual mono-source models in order to highlight the benefit to simultaneously leverage multi-source information.The use of multi-source data allows MorNet to exploit the complementarity of the three input sources and improve its performance from an f1-score of about 41.6 using a single source to 52.8 using three sources.MorNet provides a first-order moraine map through its ability to identify well-defined moraines.Consequently,MorNet can identify areas likely to contain moraines and intends to be used as a tool by experts to facilitate and support large-scale mapping.
基金by the French General Directorate for Food(DGAL),the French Agricultural Research Centre for International Development(CIRAD)and the SONGES Project(FEDER and Occitanie)supported by the French National Research Agency under the Investments for the Future Program,referred to as ANR-16-CONV-0004.by EU grant 874850 MOOD and is catalogued as MOOD010.
文摘Event-based surveillance systems are at the crossroads of human and animal(and plant and ecosystem)health,epidemiology,statistics,and informatics.Thus,their deployment faces many challenges specific to each domain and their intersections,such as relations among automation,artificial intelligence,and expertise.In this context,ourwork pertins to the extraction of epidemiological events in textual data(i.e.news)by unsupervised methods.We define the event extraction task as detecting pairs of epidemiological entities(e.g.a disease name and location).The quality of the ranked lists of pairs was evaluated using specific ranking evaluation metrics.We used a publicly available annotated corpus of 438 documents(i.e.news articles)related to animal disease events.The statistical approach was able to detect event-related pairs of epidemiological features with a good trade-off between precision and recall.Our results showed that using a window of words outperformed document-based and sentence-based approaches,while reducing the probability of detecting false pairs.Our results indicated that Mutual Information was less adapted than the Dice coefficient for ranking pairs of features in the event extraction framework.We believe that Mutual Information would be more relevant for rare pair detection(i.e.weak signals),but requires higher manual curation to avoid false positive extraction pairs.Moreover,generalising the country-level spatial features enabled better discrimination(i.e.ranking)of relevant disease-location pairs for event extraction.
基金EU grant 874850 MOOD and is catalogued as MOOD009the French General Directorate for Food(DGAL),the French Agricultural Research Centre for International Development(CIRAD),the SONGES Project(FEDER and Occitanie),and the French National Research Agency under the Investments for the Future Program,referred to as ANR-16-CONV-0004(#DigitAg).
文摘Several internet-based surveillance systems have been created to monitor the web for animal health surveillance.These systems collect a large amount of news dealing with outbreaks related to animal diseases.Automatically identifying news articles that describe the same outbreak event is a key step to quickly detect relevant epidemiological information while alleviating manual curation of news content.This paper addresses the task of retrieving news articles that are related in epidemiological terms.We tackle this issue using text mining and feature fusion methods.The main objective of this paper is to identify a textual representation in which two articles that share the same epidemiological content are close.We compared two types of representations(i.e.,features)to represent the documents:(i)morphosyntactic features(i.e.,selection and transformation of all terms from the news,based on classical textual processing steps)and(ii)lexicosemantic features(i.e.,selection,transformation and fusion of epidemiological terms including diseases,hosts,locations and dates).We compared two types of term weighing(i.e.,Boolean and TF-IDF)for both representations.To combine and transform lexicosemantic features,we compared two data fusion techniques(i.e.,early fusion and late fusion)and the effect of features generalisation,while evaluating the relative importance of each type of feature.We conducted our analysis using a corpus composed of a subset of news articles in English related to animal disease outbreaks.Our results showed that the combination of relevant lexicosemantic(epidemiological)features using fusion methods improves classical morphosyntactic representation in the context of disease-related news retrieval.The lexicosemantic representation based on TF-IDF and feature generalisation(F-measure=0.92,r-precision=0.58)outperformed the morphosyntactic representation(F-measure=0.89,r-precision=0.45),while reducing the features space.Converting the features into lower granular features(i.e.,generalisation)contributed to improving the results of the lexicosemantic representation.Our results showed no difference between the early and late fusion approaches.Temporal features performed poorly on their own.Conversely,spatial features were the most discriminative features,highlighting the need for robust methods for spatial entity extraction,disambiguation and representation in internet-based surveillance systems.
